Network coverage in Respondent Driven Sampling

We evaluate Respondent Driven Sampling (RDS)’s claim of population representation with empirical data by exploring the network coverage in a RDS study of female sex workers (FSW) in China. We take advantage of unique information on the social networks of FSW obtained from two overlapping studies of FSWs --RDS and a venue-based sampling approach (PLACE) -- and use an exponential random graph modeling (ERGM) framework from local networks to construct the likely network from which our observed RDS is drawn. We then run recruitment chains over this simulated population and produce a sample with characteristics consistent with the observed RDS. We estimate population coverage rates by comparing population proportions and RDS sample proportions. We discuss the results in light of (a) potential estimation improvements implicit in network information, (b) strategies for improving coverage rates, and (c) multiple sources of potential variability in coverage.